kanibus-kanibus/models

🧠 Kanibus Models Directory

📁 Model Storage Strategy

This directory is reserved for custom and specialized models that are not automatically downloaded by the standard libraries.

---

🎯 Current Status: EMPTY (By Design)

Kanibus v1.0.0 uses external models that are automatically downloaded:

🔄 Auto-Downloaded Models

• MediaPipe Face Mesh - 468 landmarks + iris detection
• Hugging Face Transformers - DPT depth estimation models
• Ultralytics YOLO - Object detection and pose estimation
• Segment Anything (SAM) - Advanced segmentation
• PyTorch Vision Models - Various computer vision models

These models are cached in standard locations:

• ~/.mediapipe/ - MediaPipe models
• ~/.cache/huggingface/ - Transformer models
• ~/.cache/torch/hub/ - PyTorch Hub models
• ~/.ultralytics/ - YOLO models

---

🔮 Future Use Cases

This directory is prepared for:

📁 Custom Models (v1.1+)

models/
├── custom_eye_tracker_v2.pth      # Fine-tuned eye tracking
├── medical_gaze_analyzer.onnx     # Medical applications  
├── gaming_optimized_tracker.trt   # Gaming optimizations
├── emotion_micro_expressions.pth  # Advanced emotion AI
└── offline_models/                # Offline deployment models
    ├── face_mesh_offline.tflite
    ├── depth_estimation_lite.onnx
    └── pose_detection_mobile.trt

🎯 Specialized Models

• Domain-specific - Medical, gaming, research variants
• Performance-optimized - TensorRT, ONNX, TensorFlow Lite
• Offline variants - For air-gapped environments
• Fine-tuned models - Trained on specific datasets
• Quantized models - For edge deployment

⚡ Performance Models

• TensorRT engines - NVIDIA GPU optimization
• ONNX models - Cross-platform inference
• TensorFlow Lite - Mobile/edge deployment
• CoreML models - Apple Silicon optimization

---

📋 Model Requirements

When adding custom models:

📝 Documentation

• Add model description to this README
• Include performance benchmarks
• Document input/output specifications
• Provide usage examples

🔧 Code Integration

• Add model loading in relevant node files
• Implement fallback to auto-downloaded models
• Add model validation and error handling
• Update configuration options

📊 Performance

• Benchmark against baseline models
• Test memory usage and speed
• Validate accuracy metrics
• Check multi-GPU compatibility

---

🚀 Adding Your Own Models

1. Place Model Files

# Copy your model files
cp your_model.pth models/
cp your_model.onnx models/

2. Update Node Code

# In your node file
model_path = os.path.join(os.path.dirname(__file__), "../models/your_model.pth")
if os.path.exists(model_path):
    custom_model = torch.load(model_path)
else:
    # Fallback to auto-downloaded model
    custom_model = load_default_model()

3. Update Documentation

• Add to this README
• Update node documentation
• Include in CHANGELOG.md

---

📞 Support

For custom model integration:

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Email: staffytech@proton.me

---

This directory structure is prepared for future extensibility while keeping v1.0.0 simple and efficient.